2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_ioctl.h"
36 #include "xfs_trace.h"
38 #include "xfs_icache.h"
41 #include <linux/dcache.h>
42 #include <linux/falloc.h>
43 #include <linux/pagevec.h>
44 #include <linux/backing-dev.h>
46 static const struct vm_operations_struct xfs_file_vm_ops;
49 * Locking primitives for read and write IO paths to ensure we consistently use
50 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
57 if (type & XFS_IOLOCK_EXCL)
58 mutex_lock(&VFS_I(ip)->i_mutex);
67 xfs_iunlock(ip, type);
68 if (type & XFS_IOLOCK_EXCL)
69 mutex_unlock(&VFS_I(ip)->i_mutex);
77 xfs_ilock_demote(ip, type);
78 if (type & XFS_IOLOCK_EXCL)
79 mutex_unlock(&VFS_I(ip)->i_mutex);
83 * xfs_iozero clears the specified range supplied via the page cache (except in
84 * the DAX case). Writes through the page cache will allocate blocks over holes,
85 * though the callers usually map the holes first and avoid them. If a block is
86 * not completely zeroed, then it will be read from disk before being partially
89 * In the DAX case, we can just directly write to the underlying pages. This
90 * will not allocate blocks, but will avoid holes and unwritten extents and so
91 * not do unnecessary work.
95 struct xfs_inode *ip, /* inode */
96 loff_t pos, /* offset in file */
97 size_t count) /* size of data to zero */
100 struct address_space *mapping;
104 mapping = VFS_I(ip)->i_mapping;
106 unsigned offset, bytes;
109 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
110 bytes = PAGE_CACHE_SIZE - offset;
114 if (IS_DAX(VFS_I(ip))) {
115 status = dax_zero_page_range(VFS_I(ip), pos, bytes,
116 xfs_get_blocks_direct);
120 status = pagecache_write_begin(NULL, mapping, pos, bytes,
121 AOP_FLAG_UNINTERRUPTIBLE,
126 zero_user(page, offset, bytes);
128 status = pagecache_write_end(NULL, mapping, pos, bytes,
129 bytes, page, fsdata);
130 WARN_ON(status <= 0); /* can't return less than zero! */
141 xfs_update_prealloc_flags(
142 struct xfs_inode *ip,
143 enum xfs_prealloc_flags flags)
145 struct xfs_trans *tp;
148 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
149 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
151 xfs_trans_cancel(tp);
155 xfs_ilock(ip, XFS_ILOCK_EXCL);
156 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
158 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
159 ip->i_d.di_mode &= ~S_ISUID;
160 if (ip->i_d.di_mode & S_IXGRP)
161 ip->i_d.di_mode &= ~S_ISGID;
162 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
165 if (flags & XFS_PREALLOC_SET)
166 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
167 if (flags & XFS_PREALLOC_CLEAR)
168 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
170 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
171 if (flags & XFS_PREALLOC_SYNC)
172 xfs_trans_set_sync(tp);
173 return xfs_trans_commit(tp);
177 * Fsync operations on directories are much simpler than on regular files,
178 * as there is no file data to flush, and thus also no need for explicit
179 * cache flush operations, and there are no non-transaction metadata updates
180 * on directories either.
189 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
190 struct xfs_mount *mp = ip->i_mount;
193 trace_xfs_dir_fsync(ip);
195 xfs_ilock(ip, XFS_ILOCK_SHARED);
196 if (xfs_ipincount(ip))
197 lsn = ip->i_itemp->ili_last_lsn;
198 xfs_iunlock(ip, XFS_ILOCK_SHARED);
202 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
212 struct inode *inode = file->f_mapping->host;
213 struct xfs_inode *ip = XFS_I(inode);
214 struct xfs_mount *mp = ip->i_mount;
219 trace_xfs_file_fsync(ip);
221 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
225 if (XFS_FORCED_SHUTDOWN(mp))
228 xfs_iflags_clear(ip, XFS_ITRUNCATED);
230 if (mp->m_flags & XFS_MOUNT_BARRIER) {
232 * If we have an RT and/or log subvolume we need to make sure
233 * to flush the write cache the device used for file data
234 * first. This is to ensure newly written file data make
235 * it to disk before logging the new inode size in case of
236 * an extending write.
238 if (XFS_IS_REALTIME_INODE(ip))
239 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
240 else if (mp->m_logdev_targp != mp->m_ddev_targp)
241 xfs_blkdev_issue_flush(mp->m_ddev_targp);
245 * All metadata updates are logged, which means that we just have
246 * to flush the log up to the latest LSN that touched the inode.
248 xfs_ilock(ip, XFS_ILOCK_SHARED);
249 if (xfs_ipincount(ip)) {
251 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
252 lsn = ip->i_itemp->ili_last_lsn;
254 xfs_iunlock(ip, XFS_ILOCK_SHARED);
257 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
260 * If we only have a single device, and the log force about was
261 * a no-op we might have to flush the data device cache here.
262 * This can only happen for fdatasync/O_DSYNC if we were overwriting
263 * an already allocated file and thus do not have any metadata to
266 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
267 mp->m_logdev_targp == mp->m_ddev_targp &&
268 !XFS_IS_REALTIME_INODE(ip) &&
270 xfs_blkdev_issue_flush(mp->m_ddev_targp);
280 struct file *file = iocb->ki_filp;
281 struct inode *inode = file->f_mapping->host;
282 struct xfs_inode *ip = XFS_I(inode);
283 struct xfs_mount *mp = ip->i_mount;
284 size_t size = iov_iter_count(to);
288 loff_t pos = iocb->ki_pos;
290 XFS_STATS_INC(xs_read_calls);
292 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
293 ioflags |= XFS_IO_ISDIRECT;
294 if (file->f_mode & FMODE_NOCMTIME)
295 ioflags |= XFS_IO_INVIS;
297 if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) {
298 xfs_buftarg_t *target =
299 XFS_IS_REALTIME_INODE(ip) ?
300 mp->m_rtdev_targp : mp->m_ddev_targp;
301 /* DIO must be aligned to device logical sector size */
302 if ((pos | size) & target->bt_logical_sectormask) {
303 if (pos == i_size_read(inode))
309 n = mp->m_super->s_maxbytes - pos;
310 if (n <= 0 || size == 0)
316 if (XFS_FORCED_SHUTDOWN(mp))
320 * Locking is a bit tricky here. If we take an exclusive lock
321 * for direct IO, we effectively serialise all new concurrent
322 * read IO to this file and block it behind IO that is currently in
323 * progress because IO in progress holds the IO lock shared. We only
324 * need to hold the lock exclusive to blow away the page cache, so
325 * only take lock exclusively if the page cache needs invalidation.
326 * This allows the normal direct IO case of no page cache pages to
327 * proceeed concurrently without serialisation.
329 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
330 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
331 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
332 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
334 if (inode->i_mapping->nrpages) {
335 ret = filemap_write_and_wait_range(
336 VFS_I(ip)->i_mapping,
337 pos, pos + size - 1);
339 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
344 * Invalidate whole pages. This can return an error if
345 * we fail to invalidate a page, but this should never
346 * happen on XFS. Warn if it does fail.
348 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
349 pos >> PAGE_CACHE_SHIFT,
350 (pos + size - 1) >> PAGE_CACHE_SHIFT);
354 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
357 trace_xfs_file_read(ip, size, pos, ioflags);
359 ret = generic_file_read_iter(iocb, to);
361 XFS_STATS_ADD(xs_read_bytes, ret);
363 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
368 xfs_file_splice_read(
371 struct pipe_inode_info *pipe,
375 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
379 XFS_STATS_INC(xs_read_calls);
381 if (infilp->f_mode & FMODE_NOCMTIME)
382 ioflags |= XFS_IO_INVIS;
384 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
387 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
389 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
391 /* for dax, we need to avoid the page cache */
392 if (IS_DAX(VFS_I(ip)))
393 ret = default_file_splice_read(infilp, ppos, pipe, count, flags);
395 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
397 XFS_STATS_ADD(xs_read_bytes, ret);
399 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
404 * This routine is called to handle zeroing any space in the last block of the
405 * file that is beyond the EOF. We do this since the size is being increased
406 * without writing anything to that block and we don't want to read the
407 * garbage on the disk.
409 STATIC int /* error (positive) */
411 struct xfs_inode *ip,
416 struct xfs_mount *mp = ip->i_mount;
417 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
418 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
422 struct xfs_bmbt_irec imap;
424 xfs_ilock(ip, XFS_ILOCK_EXCL);
425 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
426 xfs_iunlock(ip, XFS_ILOCK_EXCL);
433 * If the block underlying isize is just a hole, then there
434 * is nothing to zero.
436 if (imap.br_startblock == HOLESTARTBLOCK)
439 zero_len = mp->m_sb.sb_blocksize - zero_offset;
440 if (isize + zero_len > offset)
441 zero_len = offset - isize;
443 return xfs_iozero(ip, isize, zero_len);
447 * Zero any on disk space between the current EOF and the new, larger EOF.
449 * This handles the normal case of zeroing the remainder of the last block in
450 * the file and the unusual case of zeroing blocks out beyond the size of the
451 * file. This second case only happens with fixed size extents and when the
452 * system crashes before the inode size was updated but after blocks were
455 * Expects the iolock to be held exclusive, and will take the ilock internally.
457 int /* error (positive) */
459 struct xfs_inode *ip,
460 xfs_off_t offset, /* starting I/O offset */
461 xfs_fsize_t isize, /* current inode size */
464 struct xfs_mount *mp = ip->i_mount;
465 xfs_fileoff_t start_zero_fsb;
466 xfs_fileoff_t end_zero_fsb;
467 xfs_fileoff_t zero_count_fsb;
468 xfs_fileoff_t last_fsb;
469 xfs_fileoff_t zero_off;
470 xfs_fsize_t zero_len;
473 struct xfs_bmbt_irec imap;
475 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
476 ASSERT(offset > isize);
479 * First handle zeroing the block on which isize resides.
481 * We only zero a part of that block so it is handled specially.
483 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
484 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
490 * Calculate the range between the new size and the old where blocks
491 * needing to be zeroed may exist.
493 * To get the block where the last byte in the file currently resides,
494 * we need to subtract one from the size and truncate back to a block
495 * boundary. We subtract 1 in case the size is exactly on a block
498 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
499 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
500 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
501 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
502 if (last_fsb == end_zero_fsb) {
504 * The size was only incremented on its last block.
505 * We took care of that above, so just return.
510 ASSERT(start_zero_fsb <= end_zero_fsb);
511 while (start_zero_fsb <= end_zero_fsb) {
513 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
515 xfs_ilock(ip, XFS_ILOCK_EXCL);
516 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
518 xfs_iunlock(ip, XFS_ILOCK_EXCL);
524 if (imap.br_state == XFS_EXT_UNWRITTEN ||
525 imap.br_startblock == HOLESTARTBLOCK) {
526 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
527 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
532 * There are blocks we need to zero.
534 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
535 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
537 if ((zero_off + zero_len) > offset)
538 zero_len = offset - zero_off;
540 error = xfs_iozero(ip, zero_off, zero_len);
545 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
546 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
553 * Common pre-write limit and setup checks.
555 * Called with the iolocked held either shared and exclusive according to
556 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
557 * if called for a direct write beyond i_size.
560 xfs_file_aio_write_checks(
562 struct iov_iter *from,
565 struct file *file = iocb->ki_filp;
566 struct inode *inode = file->f_mapping->host;
567 struct xfs_inode *ip = XFS_I(inode);
569 size_t count = iov_iter_count(from);
572 error = generic_write_checks(iocb, from);
576 error = xfs_break_layouts(inode, iolock, true);
581 * If the offset is beyond the size of the file, we need to zero any
582 * blocks that fall between the existing EOF and the start of this
583 * write. If zeroing is needed and we are currently holding the
584 * iolock shared, we need to update it to exclusive which implies
585 * having to redo all checks before.
587 * We need to serialise against EOF updates that occur in IO
588 * completions here. We want to make sure that nobody is changing the
589 * size while we do this check until we have placed an IO barrier (i.e.
590 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
591 * The spinlock effectively forms a memory barrier once we have the
592 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
593 * and hence be able to correctly determine if we need to run zeroing.
595 spin_lock(&ip->i_flags_lock);
596 if (iocb->ki_pos > i_size_read(inode)) {
599 spin_unlock(&ip->i_flags_lock);
600 if (*iolock == XFS_IOLOCK_SHARED) {
601 xfs_rw_iunlock(ip, *iolock);
602 *iolock = XFS_IOLOCK_EXCL;
603 xfs_rw_ilock(ip, *iolock);
604 iov_iter_reexpand(from, count);
607 * We now have an IO submission barrier in place, but
608 * AIO can do EOF updates during IO completion and hence
609 * we now need to wait for all of them to drain. Non-AIO
610 * DIO will have drained before we are given the
611 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
614 inode_dio_wait(inode);
617 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
621 spin_unlock(&ip->i_flags_lock);
624 * Updating the timestamps will grab the ilock again from
625 * xfs_fs_dirty_inode, so we have to call it after dropping the
626 * lock above. Eventually we should look into a way to avoid
627 * the pointless lock roundtrip.
629 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
630 error = file_update_time(file);
636 * If we're writing the file then make sure to clear the setuid and
637 * setgid bits if the process is not being run by root. This keeps
638 * people from modifying setuid and setgid binaries.
640 return file_remove_suid(file);
644 * xfs_file_dio_aio_write - handle direct IO writes
646 * Lock the inode appropriately to prepare for and issue a direct IO write.
647 * By separating it from the buffered write path we remove all the tricky to
648 * follow locking changes and looping.
650 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
651 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
652 * pages are flushed out.
654 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
655 * allowing them to be done in parallel with reads and other direct IO writes.
656 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
657 * needs to do sub-block zeroing and that requires serialisation against other
658 * direct IOs to the same block. In this case we need to serialise the
659 * submission of the unaligned IOs so that we don't get racing block zeroing in
660 * the dio layer. To avoid the problem with aio, we also need to wait for
661 * outstanding IOs to complete so that unwritten extent conversion is completed
662 * before we try to map the overlapping block. This is currently implemented by
663 * hitting it with a big hammer (i.e. inode_dio_wait()).
665 * Returns with locks held indicated by @iolock and errors indicated by
666 * negative return values.
669 xfs_file_dio_aio_write(
671 struct iov_iter *from)
673 struct file *file = iocb->ki_filp;
674 struct address_space *mapping = file->f_mapping;
675 struct inode *inode = mapping->host;
676 struct xfs_inode *ip = XFS_I(inode);
677 struct xfs_mount *mp = ip->i_mount;
679 int unaligned_io = 0;
681 size_t count = iov_iter_count(from);
682 loff_t pos = iocb->ki_pos;
684 struct iov_iter data;
685 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
686 mp->m_rtdev_targp : mp->m_ddev_targp;
688 /* DIO must be aligned to device logical sector size */
689 if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask))
692 /* "unaligned" here means not aligned to a filesystem block */
693 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
697 * We don't need to take an exclusive lock unless there page cache needs
698 * to be invalidated or unaligned IO is being executed. We don't need to
699 * consider the EOF extension case here because
700 * xfs_file_aio_write_checks() will relock the inode as necessary for
701 * EOF zeroing cases and fill out the new inode size as appropriate.
703 if (unaligned_io || mapping->nrpages)
704 iolock = XFS_IOLOCK_EXCL;
706 iolock = XFS_IOLOCK_SHARED;
707 xfs_rw_ilock(ip, iolock);
710 * Recheck if there are cached pages that need invalidate after we got
711 * the iolock to protect against other threads adding new pages while
712 * we were waiting for the iolock.
714 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
715 xfs_rw_iunlock(ip, iolock);
716 iolock = XFS_IOLOCK_EXCL;
717 xfs_rw_ilock(ip, iolock);
720 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
723 count = iov_iter_count(from);
725 end = pos + count - 1;
727 if (mapping->nrpages) {
728 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
733 * Invalidate whole pages. This can return an error if
734 * we fail to invalidate a page, but this should never
735 * happen on XFS. Warn if it does fail.
737 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
738 pos >> PAGE_CACHE_SHIFT,
739 end >> PAGE_CACHE_SHIFT);
745 * If we are doing unaligned IO, wait for all other IO to drain,
746 * otherwise demote the lock if we had to flush cached pages
749 inode_dio_wait(inode);
750 else if (iolock == XFS_IOLOCK_EXCL) {
751 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
752 iolock = XFS_IOLOCK_SHARED;
755 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
758 ret = mapping->a_ops->direct_IO(iocb, &data, pos);
760 /* see generic_file_direct_write() for why this is necessary */
761 if (mapping->nrpages) {
762 invalidate_inode_pages2_range(mapping,
763 pos >> PAGE_CACHE_SHIFT,
764 end >> PAGE_CACHE_SHIFT);
769 iov_iter_advance(from, ret);
773 xfs_rw_iunlock(ip, iolock);
776 * No fallback to buffered IO on errors for XFS. DAX can result in
777 * partial writes, but direct IO will either complete fully or fail.
779 ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip)));
784 xfs_file_buffered_aio_write(
786 struct iov_iter *from)
788 struct file *file = iocb->ki_filp;
789 struct address_space *mapping = file->f_mapping;
790 struct inode *inode = mapping->host;
791 struct xfs_inode *ip = XFS_I(inode);
794 int iolock = XFS_IOLOCK_EXCL;
796 xfs_rw_ilock(ip, iolock);
798 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
802 /* We can write back this queue in page reclaim */
803 current->backing_dev_info = inode_to_bdi(inode);
806 trace_xfs_file_buffered_write(ip, iov_iter_count(from),
808 ret = generic_perform_write(file, from, iocb->ki_pos);
809 if (likely(ret >= 0))
813 * If we hit a space limit, try to free up some lingering preallocated
814 * space before returning an error. In the case of ENOSPC, first try to
815 * write back all dirty inodes to free up some of the excess reserved
816 * metadata space. This reduces the chances that the eofblocks scan
817 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
818 * also behaves as a filter to prevent too many eofblocks scans from
819 * running at the same time.
821 if (ret == -EDQUOT && !enospc) {
822 enospc = xfs_inode_free_quota_eofblocks(ip);
825 } else if (ret == -ENOSPC && !enospc) {
826 struct xfs_eofblocks eofb = {0};
829 xfs_flush_inodes(ip->i_mount);
830 eofb.eof_scan_owner = ip->i_ino; /* for locking */
831 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
832 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
836 current->backing_dev_info = NULL;
838 xfs_rw_iunlock(ip, iolock);
845 struct iov_iter *from)
847 struct file *file = iocb->ki_filp;
848 struct address_space *mapping = file->f_mapping;
849 struct inode *inode = mapping->host;
850 struct xfs_inode *ip = XFS_I(inode);
852 size_t ocount = iov_iter_count(from);
854 XFS_STATS_INC(xs_write_calls);
859 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
862 if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode))
863 ret = xfs_file_dio_aio_write(iocb, from);
865 ret = xfs_file_buffered_aio_write(iocb, from);
870 XFS_STATS_ADD(xs_write_bytes, ret);
872 /* Handle various SYNC-type writes */
873 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
880 #define XFS_FALLOC_FL_SUPPORTED \
881 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
882 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
883 FALLOC_FL_INSERT_RANGE)
892 struct inode *inode = file_inode(file);
893 struct xfs_inode *ip = XFS_I(inode);
895 enum xfs_prealloc_flags flags = 0;
896 uint iolock = XFS_IOLOCK_EXCL;
898 bool do_file_insert = 0;
900 if (!S_ISREG(inode->i_mode))
902 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
905 xfs_ilock(ip, iolock);
906 error = xfs_break_layouts(inode, &iolock, false);
910 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
911 iolock |= XFS_MMAPLOCK_EXCL;
913 if (mode & FALLOC_FL_PUNCH_HOLE) {
914 error = xfs_free_file_space(ip, offset, len);
917 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
918 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
920 if (offset & blksize_mask || len & blksize_mask) {
926 * There is no need to overlap collapse range with EOF,
927 * in which case it is effectively a truncate operation
929 if (offset + len >= i_size_read(inode)) {
934 new_size = i_size_read(inode) - len;
936 error = xfs_collapse_file_space(ip, offset, len);
939 } else if (mode & FALLOC_FL_INSERT_RANGE) {
940 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
942 new_size = i_size_read(inode) + len;
943 if (offset & blksize_mask || len & blksize_mask) {
948 /* check the new inode size does not wrap through zero */
949 if (new_size > inode->i_sb->s_maxbytes) {
954 /* Offset should be less than i_size */
955 if (offset >= i_size_read(inode)) {
961 flags |= XFS_PREALLOC_SET;
963 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
964 offset + len > i_size_read(inode)) {
965 new_size = offset + len;
966 error = inode_newsize_ok(inode, new_size);
971 if (mode & FALLOC_FL_ZERO_RANGE)
972 error = xfs_zero_file_space(ip, offset, len);
974 error = xfs_alloc_file_space(ip, offset, len,
980 if (file->f_flags & O_DSYNC)
981 flags |= XFS_PREALLOC_SYNC;
983 error = xfs_update_prealloc_flags(ip, flags);
987 /* Change file size if needed */
991 iattr.ia_valid = ATTR_SIZE;
992 iattr.ia_size = new_size;
993 error = xfs_setattr_size(ip, &iattr);
999 * Perform hole insertion now that the file size has been
1000 * updated so that if we crash during the operation we don't
1001 * leave shifted extents past EOF and hence losing access to
1002 * the data that is contained within them.
1005 error = xfs_insert_file_space(ip, offset, len);
1008 xfs_iunlock(ip, iolock);
1015 struct inode *inode,
1018 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1020 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1027 struct inode *inode,
1030 struct xfs_inode *ip = XFS_I(inode);
1034 error = xfs_file_open(inode, file);
1039 * If there are any blocks, read-ahead block 0 as we're almost
1040 * certain to have the next operation be a read there.
1042 mode = xfs_ilock_data_map_shared(ip);
1043 if (ip->i_d.di_nextents > 0)
1044 xfs_dir3_data_readahead(ip, 0, -1);
1045 xfs_iunlock(ip, mode);
1051 struct inode *inode,
1054 return xfs_release(XFS_I(inode));
1060 struct dir_context *ctx)
1062 struct inode *inode = file_inode(file);
1063 xfs_inode_t *ip = XFS_I(inode);
1067 * The Linux API doesn't pass down the total size of the buffer
1068 * we read into down to the filesystem. With the filldir concept
1069 * it's not needed for correct information, but the XFS dir2 leaf
1070 * code wants an estimate of the buffer size to calculate it's
1071 * readahead window and size the buffers used for mapping to
1074 * Try to give it an estimate that's good enough, maybe at some
1075 * point we can change the ->readdir prototype to include the
1076 * buffer size. For now we use the current glibc buffer size.
1078 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1080 return xfs_readdir(ip, ctx, bufsize);
1084 * This type is designed to indicate the type of offset we would like
1085 * to search from page cache for xfs_seek_hole_data().
1093 * Lookup the desired type of offset from the given page.
1095 * On success, return true and the offset argument will point to the
1096 * start of the region that was found. Otherwise this function will
1097 * return false and keep the offset argument unchanged.
1100 xfs_lookup_buffer_offset(
1105 loff_t lastoff = page_offset(page);
1107 struct buffer_head *bh, *head;
1109 bh = head = page_buffers(page);
1112 * Unwritten extents that have data in the page
1113 * cache covering them can be identified by the
1114 * BH_Unwritten state flag. Pages with multiple
1115 * buffers might have a mix of holes, data and
1116 * unwritten extents - any buffer with valid
1117 * data in it should have BH_Uptodate flag set
1120 if (buffer_unwritten(bh) ||
1121 buffer_uptodate(bh)) {
1122 if (type == DATA_OFF)
1125 if (type == HOLE_OFF)
1133 lastoff += bh->b_size;
1134 } while ((bh = bh->b_this_page) != head);
1140 * This routine is called to find out and return a data or hole offset
1141 * from the page cache for unwritten extents according to the desired
1142 * type for xfs_seek_hole_data().
1144 * The argument offset is used to tell where we start to search from the
1145 * page cache. Map is used to figure out the end points of the range to
1148 * Return true if the desired type of offset was found, and the argument
1149 * offset is filled with that address. Otherwise, return false and keep
1153 xfs_find_get_desired_pgoff(
1154 struct inode *inode,
1155 struct xfs_bmbt_irec *map,
1159 struct xfs_inode *ip = XFS_I(inode);
1160 struct xfs_mount *mp = ip->i_mount;
1161 struct pagevec pvec;
1165 loff_t startoff = *offset;
1166 loff_t lastoff = startoff;
1169 pagevec_init(&pvec, 0);
1171 index = startoff >> PAGE_CACHE_SHIFT;
1172 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1173 end = endoff >> PAGE_CACHE_SHIFT;
1179 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1180 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1183 * No page mapped into given range. If we are searching holes
1184 * and if this is the first time we got into the loop, it means
1185 * that the given offset is landed in a hole, return it.
1187 * If we have already stepped through some block buffers to find
1188 * holes but they all contains data. In this case, the last
1189 * offset is already updated and pointed to the end of the last
1190 * mapped page, if it does not reach the endpoint to search,
1191 * that means there should be a hole between them.
1193 if (nr_pages == 0) {
1194 /* Data search found nothing */
1195 if (type == DATA_OFF)
1198 ASSERT(type == HOLE_OFF);
1199 if (lastoff == startoff || lastoff < endoff) {
1207 * At lease we found one page. If this is the first time we
1208 * step into the loop, and if the first page index offset is
1209 * greater than the given search offset, a hole was found.
1211 if (type == HOLE_OFF && lastoff == startoff &&
1212 lastoff < page_offset(pvec.pages[0])) {
1217 for (i = 0; i < nr_pages; i++) {
1218 struct page *page = pvec.pages[i];
1222 * At this point, the page may be truncated or
1223 * invalidated (changing page->mapping to NULL),
1224 * or even swizzled back from swapper_space to tmpfs
1225 * file mapping. However, page->index will not change
1226 * because we have a reference on the page.
1228 * Searching done if the page index is out of range.
1229 * If the current offset is not reaches the end of
1230 * the specified search range, there should be a hole
1233 if (page->index > end) {
1234 if (type == HOLE_OFF && lastoff < endoff) {
1243 * Page truncated or invalidated(page->mapping == NULL).
1244 * We can freely skip it and proceed to check the next
1247 if (unlikely(page->mapping != inode->i_mapping)) {
1252 if (!page_has_buffers(page)) {
1257 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1260 * The found offset may be less than the start
1261 * point to search if this is the first time to
1264 *offset = max_t(loff_t, startoff, b_offset);
1270 * We either searching data but nothing was found, or
1271 * searching hole but found a data buffer. In either
1272 * case, probably the next page contains the desired
1273 * things, update the last offset to it so.
1275 lastoff = page_offset(page) + PAGE_SIZE;
1280 * The number of returned pages less than our desired, search
1281 * done. In this case, nothing was found for searching data,
1282 * but we found a hole behind the last offset.
1284 if (nr_pages < want) {
1285 if (type == HOLE_OFF) {
1292 index = pvec.pages[i - 1]->index + 1;
1293 pagevec_release(&pvec);
1294 } while (index <= end);
1297 pagevec_release(&pvec);
1307 struct inode *inode = file->f_mapping->host;
1308 struct xfs_inode *ip = XFS_I(inode);
1309 struct xfs_mount *mp = ip->i_mount;
1310 loff_t uninitialized_var(offset);
1312 xfs_fileoff_t fsbno;
1317 if (XFS_FORCED_SHUTDOWN(mp))
1320 lock = xfs_ilock_data_map_shared(ip);
1322 isize = i_size_read(inode);
1323 if (start >= isize) {
1329 * Try to read extents from the first block indicated
1330 * by fsbno to the end block of the file.
1332 fsbno = XFS_B_TO_FSBT(mp, start);
1333 end = XFS_B_TO_FSB(mp, isize);
1336 struct xfs_bmbt_irec map[2];
1340 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1345 /* No extents at given offset, must be beyond EOF */
1351 for (i = 0; i < nmap; i++) {
1352 offset = max_t(loff_t, start,
1353 XFS_FSB_TO_B(mp, map[i].br_startoff));
1355 /* Landed in the hole we wanted? */
1356 if (whence == SEEK_HOLE &&
1357 map[i].br_startblock == HOLESTARTBLOCK)
1360 /* Landed in the data extent we wanted? */
1361 if (whence == SEEK_DATA &&
1362 (map[i].br_startblock == DELAYSTARTBLOCK ||
1363 (map[i].br_state == XFS_EXT_NORM &&
1364 !isnullstartblock(map[i].br_startblock))))
1368 * Landed in an unwritten extent, try to search
1369 * for hole or data from page cache.
1371 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1372 if (xfs_find_get_desired_pgoff(inode, &map[i],
1373 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1380 * We only received one extent out of the two requested. This
1381 * means we've hit EOF and didn't find what we are looking for.
1385 * If we were looking for a hole, set offset to
1386 * the end of the file (i.e., there is an implicit
1387 * hole at the end of any file).
1389 if (whence == SEEK_HOLE) {
1394 * If we were looking for data, it's nowhere to be found
1396 ASSERT(whence == SEEK_DATA);
1404 * Nothing was found, proceed to the next round of search
1405 * if the next reading offset is not at or beyond EOF.
1407 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1408 start = XFS_FSB_TO_B(mp, fsbno);
1409 if (start >= isize) {
1410 if (whence == SEEK_HOLE) {
1414 ASSERT(whence == SEEK_DATA);
1422 * If at this point we have found the hole we wanted, the returned
1423 * offset may be bigger than the file size as it may be aligned to
1424 * page boundary for unwritten extents. We need to deal with this
1425 * situation in particular.
1427 if (whence == SEEK_HOLE)
1428 offset = min_t(loff_t, offset, isize);
1429 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1432 xfs_iunlock(ip, lock);
1449 return generic_file_llseek(file, offset, whence);
1452 return xfs_seek_hole_data(file, offset, whence);
1459 * Locking for serialisation of IO during page faults. This results in a lock
1463 * sb_start_pagefault(vfs, freeze)
1464 * i_mmap_lock (XFS - truncate serialisation)
1466 * i_lock (XFS - extent map serialisation)
1470 * mmap()d file has taken write protection fault and is being made writable. We
1471 * can set the page state up correctly for a writable page, which means we can
1472 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1476 xfs_filemap_page_mkwrite(
1477 struct vm_area_struct *vma,
1478 struct vm_fault *vmf)
1480 struct inode *inode = file_inode(vma->vm_file);
1483 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1485 sb_start_pagefault(inode->i_sb);
1486 file_update_time(vma->vm_file);
1487 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1489 if (IS_DAX(inode)) {
1490 ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_direct,
1491 xfs_end_io_dax_write);
1493 ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks);
1494 ret = block_page_mkwrite_return(ret);
1497 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1498 sb_end_pagefault(inode->i_sb);
1505 struct vm_area_struct *vma,
1506 struct vm_fault *vmf)
1508 struct xfs_inode *ip = XFS_I(file_inode(vma->vm_file));
1511 trace_xfs_filemap_fault(ip);
1513 /* DAX can shortcut the normal fault path on write faults! */
1514 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(VFS_I(ip)))
1515 return xfs_filemap_page_mkwrite(vma, vmf);
1517 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1518 ret = filemap_fault(vma, vmf);
1519 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1524 static const struct vm_operations_struct xfs_file_vm_ops = {
1525 .fault = xfs_filemap_fault,
1526 .map_pages = filemap_map_pages,
1527 .page_mkwrite = xfs_filemap_page_mkwrite,
1533 struct vm_area_struct *vma)
1535 file_accessed(filp);
1536 vma->vm_ops = &xfs_file_vm_ops;
1537 if (IS_DAX(file_inode(filp)))
1538 vma->vm_flags |= VM_MIXEDMAP;
1542 const struct file_operations xfs_file_operations = {
1543 .llseek = xfs_file_llseek,
1544 .read_iter = xfs_file_read_iter,
1545 .write_iter = xfs_file_write_iter,
1546 .splice_read = xfs_file_splice_read,
1547 .splice_write = iter_file_splice_write,
1548 .unlocked_ioctl = xfs_file_ioctl,
1549 #ifdef CONFIG_COMPAT
1550 .compat_ioctl = xfs_file_compat_ioctl,
1552 .mmap = xfs_file_mmap,
1553 .open = xfs_file_open,
1554 .release = xfs_file_release,
1555 .fsync = xfs_file_fsync,
1556 .fallocate = xfs_file_fallocate,
1559 const struct file_operations xfs_dir_file_operations = {
1560 .open = xfs_dir_open,
1561 .read = generic_read_dir,
1562 .iterate = xfs_file_readdir,
1563 .llseek = generic_file_llseek,
1564 .unlocked_ioctl = xfs_file_ioctl,
1565 #ifdef CONFIG_COMPAT
1566 .compat_ioctl = xfs_file_compat_ioctl,
1568 .fsync = xfs_dir_fsync,